From product details and options to natural answers for your customers

Last week, a shop cut the price of its linen dress from 490,000 VND to 450,000 VND. The product page was updated, and so was the post. But nobody remembered the price list copied into the chatbot’s prompt. For three days, the bot kept quoting 490,000 VND.

There’s a worse case. Sizes S and M cost 450,000 VND, while L and XL cost 480,000 VND. A customer who wears an L asks the price, and the bot says 450,000 VND. The customer pays exactly that, and the shop is left with two choices: take the loss, or message back to apologize and risk losing the order.

Both mistakes come from the same place: the bot has memorized the shop’s products instead of looking them up every time a customer asks.

Memorize or look it up?

There are two ways for an AI to know what a shop sells.

The first is to memorize: copy the price list and product list into its prompt and hope it remembers everything. It’s quick at the start, but what the bot knows stays frozen on the day you pasted it in.

The second is to look it up: your products live in the shop’s catalog, and whenever a customer asks, the bot opens the catalog and searches it. Like a new salesperson who checks the price list before answering instead of guessing.

Situation	Memorize	Look it up
The shop changes a price	You must remember to edit the bot’s prompt too; forget, and it quotes the old price	Change the price in the catalog, and the next answer uses the new one
A product comes in several sizes and colors	Prices easily get mixed up between options	Each option has its own price, and the bot uses the one the customer picked
The shop adds new products	The prompt keeps growing and the bot misses things	Add it to the catalog and the bot can find it
A customer asks for something you don’t sell	The bot may make up an answer that sounds plausible	Nothing found, so the bot says so and suggests the closest item

Zinochat works the second way. The rest of this post shows how to make it work well for your shop.

Four things the AI does before every answer

Let’s follow a conversation on Messenger.

Minh Anh: How much is the beige linen dress?

1. Understand what the customer wants. She’s asking the price of a beige linen dress but hasn’t said her size. Customers ask in all sorts of ways, like “the long beige dress you posted yesterday” or “any beach dresses under 500k?”, and the AI still understands which product they’re after.
2. Find the right product. The AI searches the shop’s catalog and finds the “Long linen dress”, in beige and white, sizes S to XL.
3. Pick the right option. This dress has two prices depending on size. The customer hasn’t given one, so the AI states the price for each size group and asks, rather than guessing.
4. Reply naturally. Only once it has everything does the AI write its answer, with a real photo of the beige dress.

Bot: The long linen dress in beige comes in sizes S to XL. Sizes S and M are 450,000 VND, and sizes L and XL are 480,000 VND. What size do you usually wear? I’ll help you pick.

Sending photos already in your catalog costs no credits, so let the bot send one whenever a customer wants to see an item.

Many customers don’t type a name at all; they send a screenshot. When you turn on Image search, the AI looks through your catalog for items that look like the customer’s photo. This feature uses AI credits, and you’ll see the cost before you turn it on.

Prepare your catalog for lookups

The AI can only find what you’ve written down clearly. Before switching the bot on, go through your catalog with this list:

• Name products the way customers do. “Long linen dress” is easier to find than “VL-0231”. Your internal code already has its own product code field; it doesn’t belong in the name.
• Write a short, complete description. Fabric, cut, occasion. When a customer asks for a “beach dress”, the AI uses the description to know which items fit.
• Set up each option separately. Every combination of size and color, such as beige in size M, is its own option.
• Enter a price for every option. Even when several share the same price. The AI always uses the price of the exact option the customer chose.
• Use real photos. Clear, true to color. Photos are both for showing customers and for matching the photos customers send.
• Keep policies in their own place. Shipping fees and returns belong in a Knowledge Base. Size charts or wholesale price sheets as images belong in an Album. Don’t repeat them in every product description.

Guide the AI on how to sell, not what things cost

The bot’s Training section is where you write its prompt. Use it for how to sell: how to address customers, the tone, what to ask first, what never to do. Leave names, prices, sizes, and colors to the catalog.

Here’s a sample prompt you can copy and adapt for your shop:

• Only quote prices from the shop’s product catalog. Never guess a price.
• If a product comes in several sizes or colors, ask which one the customer wants before confirming the price.
• Suggest at most 3 products at a time, matching the customer’s needs and budget.
• Never make up promotions, gifts, or shipping fees that aren’t in the shop’s documents.
• If you can’t find the product a customer asks for, say so honestly, then suggest the closest item.

A small tip: a good instruction is one you can check by rereading the conversation. “Be really enthusiastic” is hard to check. “Ask for the size before quoting a price” is easy to verify at a glance.

Each bot sells only its own products

If your shop runs several pages, say one for cosmetics and one for fashion, you probably don’t want the cosmetics bot recommending dresses.

In each bot’s Products section:

• No products attached: the bot can advise on the shop’s whole catalog.
• One or more products attached: the bot only advises on the products you attached.

Removing a product from a bot doesn’t delete it. Its prices, options, and photos stay in the catalog, and other bots can still use it.

From the right answer to the right order

A correct answer only gets you halfway. The other half is closing the sale. Back to Minh Anh:

Minh Anh: I wear an L. I’ll take one, delivered to District 3.

Bot: The long linen dress in beige, size L, is 480,000 VND. Could you send me the recipient’s name, phone number, and full address so I can place your order?

Once the customer sends everything, the bot creates the order right in the conversation: the right product, the right option, the right quantity, plus the name, phone number, address, and delivery notes. The order appears immediately on the Orders page for your team to handle. You don’t need to install any other software.

To help the bot know when to advise and when to close, split your Funnel into a few stages, such as New customer → Qualified → Closing. Each stage has its own prompt. A customer hesitates and goes quiet? The bot follows up on the schedule you set for each stage, within the window the channel allows, and stops as soon as the customer or a teammate replies. Zalo Bot doesn’t support follow-ups yet.

Some things need a person to decide, like a customer asking for a special price or ordering in bulk. For those, set up a Handoff rule so the bot alerts your team or hands the conversation over to the person in charge.

Test before real customers see it

Before turning the bot on for a real page, open the bot’s Preview and play the customer yourself. Test messages never reach real customers, but they use AI credits as usual.

Five things worth trying:

1. Ask the price of an item that comes in several sizes. Does the bot ask for the size before confirming the price?
2. Ask for something the shop doesn’t sell. Does the bot say so honestly?
3. Ask by budget, such as “any dresses under 500k?”
4. Ask to see a photo of a specific item.
5. Send a photo of an item, if you’ve turned on Image search.

Each time you edit the prompt, start a new test conversation so you can see the effect of the change clearly.

If the bot’s answers aren’t right yet, here’s where to look first:

The bot is…	What to check
Not finding a product	Whether names and descriptions use the words customers use, and whether Product Search is still on in the bot’s Settings
Quoting the wrong price	Each option’s price in the catalog, and whether the prompt still contains an old price list
Recommending another page’s products	Attach products to the bot to limit what it sells
Not sending photos	Whether the product has photos, and whether Product Images is still on in the bot’s Settings
Not matching the customer’s photo	Whether Image search is on and has finished processing that product’s photos
Promising promotions	Add “never make up promotions” to the Training section

Start with your own catalog

A good sales bot doesn’t need to memorize every item. It needs a clear catalog and the habit of checking before it answers. That’s how a good salesperson works, too. For the bigger picture, read What is AI Customer Experience? Customer experience powered by AI.

See how Zinochat advises and closes sales on Messenger, Instagram, and Zalo on the sales chatbot page, or Sign up to build your catalog and try your own bot.